Deforestation in Brazilian private lands:

An empirical assessment of land use

changes within farms

David R. Heres, Ramon Arigoni Ortiz

and Anil Markandya

August 2012

BC3 WORKING PAPER SERIES

2012-09
The Basque Centre for Climate Change (BC3) is a Research Centre based in the Basque Country,
which aims at contributing to long-term research on the causes and consequences of Climate Change
in order to foster the creation of knowledge in this multidisciplinary science.

The BC3 promotes a highly-qualified team of researchers with the primary objective of achieving
excellence in research, training and dissemination. The Scientific Plan of BC3 is led by the Scientific
Director, Prof. Anil Markandya.

The core research avenues are:

Adaptation to and the impacts of climate change

Measures to mitigate the amount of climate change experienced

International Dimensions of Climate Policy

Developing and supporting research that informs climate policy in the Basque
Country

See www.bc3research.org for further details.

The BC3 Working Paper Series is available on the internet at

http://www.bc3research.org/working_papers/view.html

Enquiries (Regarding the BC3 Working Paper Series):

Roger Fouquet

Email: roger.fouquet@bc3research.org

The opinions expressed in this working paper do not necessarily reflect the position of Basque Centre
for Climate Change (BC3) as a whole.

Note: If printed, please remember to print on both sides. Also, perhaps try two pages on one side.
 Deforestation in Brazilian private lands: An empirical assessment of land
use changes within farms1

David R. Heres
Ramon Arigoni Ortiz
Anil Markandya

Abstract: Within the perspective of evolving negotiations for Reduced

Emissions from Deforestation and Degradation (REDD), the opportunity
costs of deforestation are regarded as the basis for constructing a REDD
mitigation cost curve. This paper presents a land-use model that measures
the impact of economic and physical variables on farmers decisions about
the allocation of their land among competing uses in Brazil. It is based on a
multi-output land allocation model with agricultural land as fixed and
allocatable input. For the first time, to the best of our knowledge, such a
land allocation model has been estimated that explicitly separates all main
competing uses for forest in Brazil soybeans, sugarcane and pasture. Our
results suggest a clear substitution effect between land allocation for forest
and soybeans, and for forest and pasture. The results presented here
represent crucial information for estimating the opportunity costs of
deforestation at the county level in Brazil.

Keywords: REDD; Deforestation; Brazil; Land-use model

JEL Classification: C30, C51, Q11, Q23, Q24

Cite as: Heres, D.R.; R.A. Ortiz and A. Markandya (2012) Deforestation in Brazilian private lands:
An empirical assessment of land use changes within farms. BC3 Working Paper Series 2012-09.
Basque Centre for Climate Change (BC3). Bilbao, Spain.

1
This paper was developed as part of the Brazil Carbon Market Project, a joint initiative of IPEA, BC3,
CIRED and CentroClima. We are grateful to participants of the workshop "Regulatory Aspects of Carbon
Market in Brazil: preliminary modelling outputs" for their comments and suggestions. We also thank Elena
Ojea for her help with some of the REDD studies. Remaining errors and omissions are ours. The model
presented here follows Moore and Negri (1992). Feres, Reis and Speranza (2010) applied another version of the
model to data from the 1995 Brazilian Agricultural Census. Data used in our estimations and not available at
the IBGE website were kindly provided by IPEA (Jose Gustavo Feres, Ronaldo Seroa da Motta, Eustaquio Reis,
Luiza Castro and Lilia Couto).

1
1. Introduction

According to a number of studies, climate change mitigation opportunities in Brazil have

great potential related to reducing emissions from deforestation and degradation (REDD) (e.g.
Nepstad et al., 2007; Chomitz et al., 2007; Mckinsey & Company, 2009). These reductions are
believed to be cheaper to obtain than in other sectors of the economy. Therefore, any national climate
change strategy pursued by the Brazilian government should include REDD opportunities as an
economic alternative for mitigation, as well as for preserving the countrys vast natural capital. Such
a strategy, in principle, would compensate farmers for avoiding deforestation of their lands with an
amount at least equal to the expected revenue that they would receive from using the lands for other
uses, principally agriculture or pasture. These opportunity costs of deforestation, estimated for each
municipality of Brazil, are the basis for constructing the REDD mitigation cost curve for Brazil. This
paper presents a land-use model that sheds light on the impact of economic and physical variables on
the decisions that farmers make regarding the allocation of their land among competing uses,
including forest. It is the first time, to the best of our knowledge, that such a land-use model has been
developed that includes all significant land-use types driving deforestation in Brazil, such as pasture,
soybean and sugar cane plantations. Therefore, our results represent crucial information for
estimating the opportunity costs of deforestation in Brazil.
The paper is organized as follows: section 2 describes the theoretical model that underlies our
work; section 3 presents the empirical model and discusses a few econometric issues that arise with
our data. Section 4 describes the data used in our analysis and section 5 presents our econometric
results. Section 6 presents a policy implication example and section 7 concludes.

2. Theoretical Model

Multi-crop production models have been widely applied to analyze farmers behavior and
production technologies (Shumway, 1983; Just, Zilberman and Hochman, 1983; Shumway, Pope, and
Nash, 1984; Chambers and Just, 1989; Moore and Negri, 1992; Moore, Gollehon and Carey, 1994;
Moore and Dinar, 1995). These models specify a profit function from which estimable output supply
and input demand equations are derived.2
As in previous work applying Brazilian land use data (Feres et al. 2010), the theoretical
model underlying the work presented here is adapted from the multi-output production model
(Cambers and Just, 1989). The model description below is based mainly on Moore and Negri (1992);
the interested reader will find further details in their original work. According to their theoretical
model the following three assumptions represent the essential features of agricultural production and
provide a tractable approach to the multi-output production, especially of the fixed but allocatable
input functions (i.e., land and water in their case, only the former in ours): (a) inputs are allocated to
specific crop production activities; (b) production is technically non-joint in the sense that the
allocation of inputs uniquely determines crop-specific output levels; and (c) land is a fixed input that
is allocated for different uses. Assumptions (a) and (b) enable the formation of separate restricted
profit functions for each crop, taking land allocations as given; and assumption (c) provides the source
of joint allocation when maximizing multi-crop profits.

2
Other approaches are related to the random utility maximization framework thus avoiding the choice of
functional forms for the profit functions (Parks, 1980; Lichtenberg, 1989; Wu and Segerson, 1995; Hardie and
Parks, 1997; Plantinga, 1996; Miller and Plantinga, 1999; Plantinga, Maudlin and Miller, 1999). Our approach,
however, will allow us in future work to derive the opportunity costs in public land.

2
 Farmers are assumed to allocate land and other inputs in order to maximize their profits ()
from different uses, constrained to the total amount of land available. Formally:

, (1)

where (n) is a vector of land allocated for i = 5 uses (sugar cane; soybeans; pasture; forest and "other
crops"); (p) is a vector of crop prices; (r) is a vector of other input prices (only labor in our case); (X)
is a vector of agro-climatic variables that may influence the farmers decision for allocating land (e.g.
temperature; precipitation; soil type and quality; average altitude; distance to markets); and (N) is the
total farmland available. The Lagrangian function (L) is as follows:

(2)

where () is the shadow price of land constraint. The necessary conditions for an interior solution are:

i=15 (3)

(4)

Equation (3) allocates land among crops to equate the marginal profit from each crop. The
input constraint in (4) is binding assuming an interior solution. Solving equations (3) and (4) yields
the optimal solutions to equation (2), denoted ni*(p, r, X, N).34 These represent the multi-crop
farmer's production equilibrium in land allocations.

3. Econometric Model

Normalized quadratic crop-specific profit functions are assumed in our empirical analysis due
to their appealing theoretical and empirical properties (Shumway, 1983). This form has been adopted
in previous empirical work on land-use decisions (Shumway, 1983; Moore and Negri, 1992; Moore et
al., 1994; Moore and Dinar, 1995; Feres et al., 2010) since Laus series of theoretical works (see for
example Lau, 1978). The normalized quadratic profit equations give a second order Taylors
approximation to an arbitrary functional form (i.e., flexible functional form) and impose linear
homogeneity of the profit functions in prices (i.e., increasing all input and output prices by the same
factor does not change the optimal choice and increases profit by that factor ). Furthermore, by
applying Hotellings lemma to these functions we get supply and variable input demands that are
linear in their parameters. The primary interest of our study lies on the land allocation functions,
which as shown in Moore and Negri (1992) are also linear in their parameters. Importantly, we do not
impose cross-price symmetry conditions in our econometric specifications as these do not hold in the
case of fixed allocatable inputs (Moore and Negri, 1992; n. 8). However, to ensure linear homogeneity
of the profit function in prices both p and r are relative prices with respect to that of other crops. The
five land allocation equations to be estimated are:
3
The six equations in (3) and (4) give the solutions for n 1 through n5 as functions of the input and output prices
and the agro-climatic variables.
4
Land allocations to sugarcane and soybean are zero in some municipalities. This would imply that a corner
solution is attained in some municipalities and that equation (3) is an inequality in some instances. Marginal
profits are still equal among those uses with non-zero fixed input allocations and are higher than those to which
no land was allocated (Chambers and Just, 1989).

3
 (5)

It is important to note that the price of land is not included as an explanatory variable in
equation (5). Instead, since land is a fixed input, the total land available is included as a regressor.5

Figure 1: Histograms for sugarcane, soybeans and forest areas in levels and logs

Due to the large number of municipalities in which sugarcane and soybeans are not grown
(35% and 75%, respectively), and to a certain degree where there is no forest as well (3%), ordinary
least squares (OLS) estimation of those equations would yield inconsistent estimates of the parameters
(Cameron and Trivedi, 2005). Moore and Negri (1992) solved this issue by applying a Tobit model to
their dataset. This model, however, requires the non-censored observations (i.e., those with positive
values in those land uses) to be normally distributed for estimates to be consistent. Our data do not
comply with that assumption; rather it looks as if the observations have a log normal distribution (see
Figure 1). Therefore a Tobit model for lognormal data is applied to the sugarcane, soybeans and forest
equations. Equations for other crops and pasture are estimated through OLS with dependent variable
in levels since this method does not require normality for consistency of its estimates and only a
negligible number of municipalities have no land in those uses (0.1% and 0.2%, respectively).
The Tobit specification applied to some equations implies that system estimation is
computationally complicated in our case. Furthermore, in the linear case efficiency is not improved
when regressors are identical across equations. As in a separate issue, although our data comes from
an agricultural census, it is possible that some estimation gains could be achieved by weighting
observations by the importance of agriculture output or number of farms in a county. So-called

5
By examining regressions in which either the price or the total amount of an input is included, Moore and
Dinar (1995) developed statistical tests aimed at discriminating between the fixed or variable assumption of land
and water inputs. Here, we adopt the common assumption of land being a fixed but allocatable input.

4
analytical weights can be attached when variables represent averages. However, although our price
variables are averages, our dependent variable is a sum. We leave these two issues for future work
and present in Table 3 results for the separate un-weighted regressions for the five land use categories.

4. Data
Land-use, production and sales data in our analysis were obtained from the latest agricultural
census available in Brazil.6 In this dataset tables correspond to farm data aggregated at the
municipality level. Thus, we recovered data on land allocation and production at the municipal level
(5,564 municipalities) for sugar cane; soybeans; other crops representing approximately 90% of the
rest of the agricultural production in Brazil7; pasture (planted and natural) and forest (planted and
natural).
Prices were estimated from information about the value of production divided by the quantity
produced for each commodity in every state. National prices estimates were also calculated and used
as a reference price in those municipalities where there was no production of a given crop8. For the
product "other crops", a price index was estimated as a weighted (quantities produced) average of the
representative crops grown in each state. Price for "cattle" was estimated dividing the value of abated
and sold herds by the quantities of abated and sold animals, which results in the weighted average
price from these goods. A similar procedure was used to estimate the price for forest products: we
used value divided by quantities of timber, wood for paper and wood for other uses.
Wages are the weighted average salaries paid to family and non-family workers divided by
number of workers. This is the only input price variable included in our regressions. We assume that
capital costs are similar across the country and that agro-climatic and distance variables capture some
of the remaining differences in variable costs. Climatic (average quarterly temperature and
precipitation from 1993 to 2002), altitude, and soil type data were provided by IPEA. These data
were collected for year 1995 and since then some municipalities in Brazil have been created through
emancipation. In other words, the geographical units in 2006 do not match with the geographical
units in the agro-climatic dataset. To address this, we make two assumptions related to the soil data
collected in 1995. First, municipalities that had split since then share the same soil characteristics as
those in the original municipality. Second, average municipality soil characteristics have not changed
during the decade preceding the latest agricultural census. Finally, municipalities that did not exist in
1995 were dropped from the analysis. Future work will incorporate these municipalities as
information becomes available regarding their origin.
Finally, macroeconomic data (GDP; GDP per capita; sectoral GDP; population) per
municipality were obtained from IPEADATA (www.ipeadata.gov.br). We complemented the dataset

6
Censo Agropecurio 2006, Instituto Brasileiro de Geografia e Estatstica IBGE; Sistema Automtico de
Recuperaao de Dados SIDRA www.sidra.ibge.gov.br.
7
Algodao herbceo (cotton); arroz em casca (rough rice); feijao em grao (beans) preto (black beans), de cor
(red beans), fradinho (black eyed beans) e verde (green beans); fumo em folha (tobacco leaf); mandioca
(cassava); milho em grao (corn grains) e forrageiro (fodder maize); sorgo em grao (sorghum); trigo em grao
(wheat grain); forrageiras para corte (fodder); cacao (cocoa); caf arbica e canephora (coffee); laranja
(orange).
8
By assuming a national average price as a reference price means that the farmers decision to allocate land is
purely economic. As correctly pointed out by a reviewer, the absence of production of a certain crop in a given
county may be due to agro-climatic constraints of the area, and in this case the reference price to be used in the
model should be zero. In our model, we use national prices as reference prices when the county does not
produce a certain good and control for agro-climatic characteristics by including county-specific dummies for
soil type, temperature and precipitation levels in each county.

5
with distances from each municipality to the Federal District (Brasilia) and to the nearest sea shore 9.
Tables 1 and 2 respectively describe and summarize the data used in our analysis.

Table 1: Data description

variable description
p_sugarcane (1) State-level sugar cane price (R$1000 / ton)
p_soybeans (1) State-level soybeans price (R$1000 / ton)
(1)
p_cattle State-level cattle price (R$1000 / head)
p_timber (1) State-level timber and wood price (R$1000 / 1000 m3)
wages (1) State-level average annual salary paid per family and non-family worker (R$1000)
a_sugarcane Area of the municipality used to grow sugarcane (hectares)
a_soybeans Area of the municipality used to grow soybeans (hectares)
a_pasture Area of the municipality that is pasture (hectares)
a_forest Area of the municipality that is forest (hectares)
a_crops Area of the municipality used to grow other crops (hectares)
a_municip Total area of the municipality (hectares, excluding constructed areas)
Aggdp Share of total GDP from agriculture (%) in the municipality
dist_shore Distance from municipality to the nearest shore (meters)
dist_captl Distance from municipality to Brasilia-DF (km)
temp1 1993-2002 municipality average temperature for Dec-Feb (oC)
temp2 1993-2002 municipality average temperature for Mar-May (oC)
temp3 1993-2002 municipality average temperature for Jun-Aug (oC)
temp4 1993-2002 municipality average temperature for Sep-Nov (oC)
prec1 1993-2002 municipality average precipitation for Dec-Feb (millimeters)
prec2 1993-2002 municipality average precipitation for Mar-May (millimeters)
prec3 1993-2002 municipality average precipitation for Jun-Aug (millimeters)
prec4 1993-2002 municipality average precipitation for Sep-Nov (millimeters)
Altitude Average altitude (meters) in the municipality
perc_alt2 % of municipality area between 100 and 199 meters high
perc_alt3 % of municipality area between 200 and 499 meters high
perc_alt4 % of municipality area between 500 and 799 meters high
perc_alt5 % of municipality area between 800 and 1199 meters high
perc_alt6 % of municipality area between 1200 and 1799 meters high
perc_agpot2 % of municipality classified as level 2 agricultural potential
perc_agpot3 % of municipality classified as level 3 agricultural potential
perc_agpot4 % of municipality classified as level 4 agricultural potential
perc_agpot5 % of municipality classified as level 5 agricultural potential
perc_soil1 % of municipality area with class 1 soil type
perc_soil2 % of municipality area with class 2 soil type
perc_soil3 % of municipality area with class 3 soil type
perc_soil4 % of municipality area with class 4 soil type
perc_soil5 % of municipality area with class 5 soil type
perc_soil6 % of municipality area with class 6 soil type
perc_soil7 % of municipality area with class 7 soil type
perc_soil8 % of municipality area with class 8 soil type
perc_soil9 % of municipality area with class 9 soil type
perc_soil10 % of municipality area with class 10 soil type
Note: (1) Price relative to state-level "other crops" price; all data refer to year 2006.

9
These variables intend to approximate the level of access to markets for agricultural goods, cattle and timber.
Another good proxy would be the road density in the municipality, as suggested by a reviewer. However, in the
absence of such data we only include road distance to the capital and to the nearest shore..

6
Table 2: Descriptive summary of data
Variable mean standard deviation minimum maximum
p_sugarcane 0.215 0.195 0.028 2.277
p_soybeans 1.294 1.012 0.259 6.336
p_cattle 1.478 0.681 0.363 3.444
p_timber 32.657 19.632 5.042 70.752
wages 4.933 5.834 0.385 18.938
a_sugarcane 969.33 4552.20 0.00 87337.00
a_soybeans 2741.71 13379.80 0.00 443921.00
a_pasture 30120.38 83146.24 0.00 3693974.00
a_forest 7303.04 18445.32 0.00 368493.00
a_crops 7500.59 11651.47 0.00 266541.00
a_municip 48635.05 100195.20 0.00 4026075.00
Aggdp 0.217 0.143 0.00 0.788
dist_shore 346225.60 352174.30 0.00 2494990.00
dist_captl 1392.123 623.557 0.00 4600.000
temp1 25.354 1.713 18.690 28.716
temp2 23.565 2.720 15.552 27.760
temp3 20.998 4.082 11.234 28.494
temp4 23.984 3.229 14.689 30.353
prec1 181.140 79.934 29.070 413.713
prec2 128.300 59.434 29.144 488.084
prec3 61.202 57.623 0.023 424.880
prec4 106.471 54.721 2.258 268.464
Altitude 417.697 293.224 0.00 1628.000
perc_alt1 0.148 0.320 0.00 1.00
perc_alt2 0.126 0.248 0.00 1.00
perc_alt3 0.356 0.387 0.00 1.00
perc_alt4 0.269 0.342 0.00 1.00
perc_alt5 0.093 0.220 0.00 1.00
perc_alt6 0.007 0.042 0.00 0.50
perc_agpot1 0.075 0.238 0.00 1.00
perc_agpot2 0.108 0.266 0.00 1.00
perc_agpot3 0.004 0.056 0.00 1.00
perc_agpot4 0.567 0.426 0.00 1.00
perc_agpot5 0.246 0.368 0.00 1.00
perc_soil1 0.296 0.389 0.00 1.00
perc_soil2 0.108 0.269 0.00 1.00
perc_soil3 0.008 0.081 0.00 1.00
perc_soil4 0.003 0.047 0.00 1.00
perc_soil5 0.009 0.070 0.00 1.00
perc_soil6 0.321 0.397 0.00 1.00
perc_soil7 0.034 0.159 0.00 1.00
perc_soil8 0.120 0.265 0.00 1.00
perc_soil9 0.030 0.155 0.00 1.00
perc_soil10 0.039 0.165 0.00 1.00

5. Results

Equations for sugarcane, soybeans and forest were estimated using lognormal Tobit whereas
equations for pasture and other crops using ordinary least squares. Therefore the comparison of results
across equations based on the regression coefficients is not straightforward. In order to facilitate this

7
comparison across equations we present the corresponding land allocation elasticities in Table 3
(detailed results from Tobit and OLS models are available from the authors upon request).

Table 3: Land allocation elasticities

variable sugarcane soybeans pasture forest other crops
p_sugarcane -0.049 -0.136 -0.008 0.219 -0.049
(0.075) (0.220) (0.031) (0.037)*** (0.035)
p_soybeans -0.477 1.075 0.060 -0.315 -0.160
(0.116)*** (0.363)*** (0.013)*** (0.053)*** (0.026)***
p_cattle -0.112 1.592 0.336 -1.242 -0.367
(0.239) (0.609)*** (0.046)*** (0.105)*** (0.072)***
p_timber 0.556 0.898 -0.272 0.422 0.211
(0.209)*** (0.419)** (0.077)*** (0.116)*** (0.112)*
wages 0.302 -0.647 0.055 -0.215 0.006
(0.106)*** (0.228)*** (0.038) (0.059)*** (0.055)
a_municip 0.143 0.874 1.279 0.395 0.290
(0.048)*** (0.110)*** (0.072)*** (0.110)*** (0.111)***
aggdp -0.302 1.128 -0.000 0.074 0.048
(0.085)*** (0.194)*** (0.029) (0.047) (0.048)
dist_shore 0.174 2.254 -0.017 0.066 0.127
(0.126) (0.332)*** (0.056) (0.084) (0.091)
dist_captl -0.650 -7.296 -0.193 0.343 -0.061
(0.279)** (0.738)*** (0.100)* (0.161)** (0.183)
temp1 -16.603 24.773 6.743 -2.821 -4.978
(4.453)*** (12.175)** (1.216)*** (2.122) (1.977)**
temp2 18.119 -107.048 -2.303 -6.396 -0.277
(4.811)*** (11.474)*** (1.151)** (2.136)*** (1.755)
temp3 -32.034 -2.865 1.540 6.793 -0.731
(3.604)*** (9.379) (0.710)** (1.561)*** (1.234)
temp4 33.283 55.885 -3.551 -2.312 2.146
(4.255)*** (11.769)*** (0.867)*** (1.913) (1.540)
prec1 1.417 -12.455 0.843 -2.235 -1.437
(0.454)*** (1.205)*** (0.110)*** (0.228)*** (0.196)***
prec2 -1.488 8.420 -0.480 1.014 0.900
(0.399)*** (1.040)*** (0.127)*** (0.217)*** (0.219)***
prec3 0.584 -0.525 0.130 -0.529 -0.196
(0.184)*** (0.516) (0.035)*** (0.082)*** (0.063)***
prec4 -0.961 8.007 -0.357 0.646 0.081
(0.387)** (0.983)*** (0.092)*** (0.169)*** (0.177)
altitude 0.204 2.030 -0.060 -0.066 0.162
(0.179) (0.383)*** (0.034)* (0.083) (0.058)***
perc_alt2 -0.062 0.092 0.023 0.084 -0.066
(0.043) (0.114) (0.005)*** (0.019)*** (0.013)***
perc_alt3 -0.627 -0.084 0.028 0.291 -0.230
(0.109)*** (0.297) (0.018) (0.052)*** (0.034)***
perc_alt4 -0.636 -0.365 -0.012 0.364 -0.097
(0.107)*** (0.262) (0.016) (0.052)*** (0.031)***
perc_alt5 -0.308 -0.507 0.007 0.082 -0.031
(0.052)*** (0.121)*** (0.007) (0.025)*** (0.014)**
perc_alt6 -0.040 -0.173 0.005 0.003 -0.017
(0.010)*** (0.045)*** (0.001)*** (0.004) (0.002)***

8
perc_agpot2 0.021 -0.260 0.004 0.049 -0.019
(0.034) (0.074)*** (0.006) (0.017)*** (0.011)*
perc_agpot3 -0.003 -0.044 0.004 -0.001 -0.006
(0.004) (0.026)* (0.002)* (0.003) (0.003)**
perc_agpot4 -0.515 -1.224 0.036 0.284 -0.099
(0.165)*** (0.374)*** (0.033) (0.083)*** (0.058)*
perc_agpot5 -0.088 0.240 0.029 0.021 -0.099
(0.091) (0.233) (0.013)** (0.042) (0.026)***
perc_soil1 0.860 -0.349 -0.028 -0.335 -0.123
(0.156)*** (0.461) (0.032) (0.064)*** (0.071)*
perc_soil2 0.189 0.056 -0.020 -0.023 -0.020
(0.058)*** (0.168) (0.011)* (0.024) (0.026)
perc_soil3 -0.020 0.046 0.002 -0.008 -0.005
(0.007)*** (0.015)*** (0.001) (0.003)*** (0.002)**
perc_soil4 -0.006 -0.529 0.002 -0.010 -0.005
(0.006) (0.133)*** (0.001) (0.004)*** (0.002)**
perc_soil5 0.023 -0.053 -0.000 -0.010 0.001
(0.009)*** (0.041) (0.002) (0.005)* (0.005)
perc_soil6 1.051 1.718 -0.124 -0.281 -0.005
(0.174)*** (0.500)*** (0.035)*** (0.071)*** (0.077)
perc_soil7 0.005 -0.156 -0.007 0.005 -0.007
(0.022) (0.072)** (0.003)* (0.009) (0.008)
perc_soil8 0.104 0.007 -0.031 0.013 0.004
(0.061)* (0.179) (0.013)** (0.024) (0.031)
perc_soil9 0.049 0.134 -0.014 -0.026 -0.007
(0.020)** (0.049)*** (0.004)*** (0.008)*** (0.008)
perc_soil10 -0.105 0.234 -0.002 -0.035 -0.004
(0.027)*** (0.072)*** (0.004) (0.011)*** (0.011)
pseudo-R2 0.072 0.273 0.926 0.115 0.206
N 4891 4891 4891 4891 4891
N (y=0) 1736 3691 11 125 5
Elasticities calculated at mean values from lognormal Tobit (sugarcane, soybeans and forest)
and OLS (pasture and crops). Robust standard errors in parentheses.
Significance at the 10%, 5%, and 1% level denoted by ***, ** and * respectively.

From Table 3, goodness of fit of our models is generally satisfactory, with many regressors
presenting the expected signs and significance at the usual levels. Pseudo-R-square statistics (R-
square in OLS equations) are typical of cross-sectional analysis. The positive exception is the
unusually high goodness of fit of the pasture equation, indicating that the explanatory variables in our
model explain 92.6% of the observed land allocation for pasture.
Own-price elasticities are positive and statistically significant, as expected, except for
sugarcane. These results indicate that a marginal increase in output price, all other factors being
constant, increases the land allocation of the corresponding output. The observed own-price effect is
especially large for soybeans (land allocation elasticity equal to 1.075). The impact of sugarcane
price (relative to other crops) is not significant in any equation except forest. The ineffective
performance of sugarcane price raises a number of conjectures. It can partially be explained by the
multiple use of this crop in some regions in Brazil as sugarcane can be used for ethanol and sugar
production, depending on the market prices of these commodities at the time of harvesting. This fact
may allow producers to put less importance to sugarcane prices when deciding their land allocation
for sugarcane since risks are reduced by the double use characteristic of this crop. In fact, we observe
high variability of all our price estimates, even when aggregated at the state level. This is a limitation

9
of our dataset that perhaps affects more the sugarcane equation than others. Secondly, it is likely that
endogeneity may play a negative role in our model, since we assume that prices determine land
allocation and it may be the case that for sugarcane the amount of output (and therefore the amount of
land allocated) drives the price of sugarcane. Finally, it may be the case that for sugarcane the
appropriate price to be used is the lagged-price, i.e. that farmers determine their land allocation driven
by crop prices in the previous year10.
A particularly relevant result for the objectives of our study is that price of soybeans and
cattle affect negatively and significantly land allocation for forest. However, price of timber impacts
negatively land allocation for pasture but positively all other land uses, including soybeans. This later
result suggests that cross-price effects are not symmetric in our model, as expected in the case of fixed
allocatable inputs.
The elasticities of area in a given use with respect to total land are all statistically significant
and positive, as expected since they measure the additional allocation in a specific land use given a
marginal increase in total land. Our land allocation elasticity results show that pasture and soybeans
present much higher elasticity in regard to total agricultural land than other land uses. For example,
an extra hectare of agricultural land corresponds to an increase of 0.7922 hectare of pasture. One
especially relevant and positive result of our study concerns the additive property of total land
allocation (physical constraint, equation 4). In other words, the coefficients of land constraint should
sum up to one when all land uses are accounted for.
Wages have a mixed impact in our land allocation equations. Coefficients for wages (our
input price) are not statistically significant for pasture and other crops, perhaps reflecting the lower
labor intensity of these activities, especially cattle ranching. However, these are statistically
significant for sugarcane, soybeans and forest. The negative sign of this coefficient in the soybean
equation reflects the high level of mechanization in soybean plantations in Brazil. The positive effect
of wages in land allocation for sugarcane (a labor intensive activity), may be driven by the fact that
sugarcane is mostly produced in the state of So Paulo, a highly industrialized region where wages are
certainly above the national level, in all sectors of the economy.
Finally, climate (temperature and precipitation) and soil coefficients are in general
statistically significant, confirming that these factors are important for farmers land allocation among
the different land uses. It is difficult, however, to figure out the expected signs of these coefficients
because location-specific physical variables in a multi output production model (with fixed,
allocatable inputs) measure a municipalitys comparative advantage in producing a crop rather than its
absolute advantage (Moore and Negri, 1992).

6. Policy Implications of Results for REDD

The implications of these results for a program such as REDD are not straightforward but
nevertheless we can say something of interest. The cross section of data is assumed to represent the
current equilibrium of land use in Brazil between different crops and forest. In the future, changes
in this pattern (in particular a loss of forest) will be driven (inter alia) by expected increases in the
prices of soya, cattle, sugarcane and other crops relative to forest. Hence any program that offers an
incentive to maintain standing forest in that form has to be judged relative to the incentives that may
exist to clear the forest and convert it to agriculture.

10
The lagged-price effect is likely to be true for all crops. However, the Agricultural Census is undertaken
every five years in Brazil, which makes it impossible to estimate lagged prices in such a large period difference.
It is possible, however, to estimate lagged agricultural prices by using a different data source, the Pesquisa
Agrcola Municipal PAM, which is undertaken every year. This may be an interesting improvement for future
versions of the model.

10
 It is difficult to know what the incentives are likely to be for clearance but, as an example, we
consider two: 1) an expected increase in the price of cattle of 10% and 2) an expected increase in the
price of Soya of 10%. The results of applying the elasticities in Table 3 allow us to calculate the
change in land areas. To be sure these changes depend on the baseline values of areas in the
municipality under each crop. For the purposes of illustration we take a typical area with the
average values of pasture and of each crop, as given in Table 2. We then consider three levels of
payment for the avoided deforestation in terms of US$ per ton of CO2 avoided: US$5, US$10 and
US$20. These payments need to be converted into payments per hectare of forest, which has been
done using parameters given in the notes to Table 4. The REDD payments are then compared against
the net present value of the return from the land under ranching or soya. The results of changes in land
converted are shown in that Table.
The incentives for forest clearing do indeed result in some clearance (around 9 percent in the
case of cattle prices and 6 percent in the case of soya prices). At the same time they also result in
shifts in other crops (following the results generated in this paper11). These impacts can, however, be
reduced if there is a REDD premium for holding forest as standing timber. The first set of
calculations for the price of cattle shows that a modest price of US$5 would act to reduce the amount
of deforestation by 60 percent (from 678 ha. to 272 ha.). Further increases in REDD reduce the
amount deforested, so with a US$20 payment the amount cleared is only 5 percent of what it would be
without any payment.
In the case of soya the impact of REDD is similar. The US$5 payment reduces to amount of
deforestation by about 75 percent and a payment of US$20/ton reduces it by 97 percent12.
These results can be compared with those of many others who have attempted to estimate the
impacts of different REDD programs on deforestation. In the two global models used in the Eliasch
Review (Eliasch, 2008), the carbon prices that would halve forest emissions are US$11 and US$15
per ton of CO2 respectively. In comparison, Brner and Wunder (2008), in a local-empirical study in
two states in Brazil, estimate a choke price (that would prevent all deforestation) at US$12.34 in
Mato Grosso and US$3.24 in Amazonas. Another study of the whole Brazilian Amazon (Nepstad et
al. 2007) produces much lower costs still, finding that 90 per cent of the forest emissions can be
abated for less than US$1.4/t CO2 (US$5/t C). At a local level for Brazil, Olsen and Bishop (2009),
estimate a range of values for the opportunity cost of land, depending on possible alternative uses.
These range from almost nothing for beef cattle production to US$ 2.6 tCO2e and between US$2.5
and $3.4/tCO2e for soybean production. Tree plantations present high returns, but cover only a small
area of the Amazon. As approximately 80% of recently deforested land is used for cattle ranching,
they argue that the scope for achieving cost-effective reductions in CO2e emissions through avoided
deforestation is high.
Our model allows for a somewhat different analysis. It is the first model that looks at land
use allocations in an overall profit making structure and thus offers the possibility to examine
combinations of changes in future incentives for deforestation. We also emphasize the need to look at
the problem in a dynamic context, where deforestation will be driven by expected changes in the
prices of crops. Finally, few of the researchers have incorporated in their models the wide variations
in returns within a region or country. Olsen and Bishop recognize the variability across provinces but
in our experience there are also considerable variations within a province or indeed even within a
municipality, as our data show.

11
According to the regressions the net changes in land should be zero. In practice they are not because applying
the elasticities results in non-marginal changes; we have applied a scaling adjustment factor to bring the net
changes to zero.
12
To capture the very real feature of land use (i.e. that profitability varies widely) we assume a log normal
distribution of profit per ha., with the mean values as in Table 2 and a plausible standard deviation.

11
Table 4: Impacts of a REDD Program on a Typical Municipality in Brazil

Impact of a 10% expected increase in the price of cattle

Price of CO2 Gain from REDD Forest Ha. Pasture Ha. Other land use Net Change
Under REDD R$000/Ha. Ha. Ha.
$/Ton
0 0 -678 1,012 -334 0
5 8.037 -272 606 -334 0
10 16.075 -117 451 -334 0
20 32.149 -34 369 -334 0

Impact of a 10% expected increase in the price of soybeans

Price of CO2 Gain from REDD Forest Ha. Soya Ha. Other land use Net Change
Under REDD R$000/Ha. Ha. Ha.
$/Ton
0 0 -396 295 101 0
5 8.037 -98 -3 101 0
10 16.075 -37 -64 101 0
20 32.149 -10 -91 101 0

Assumptions and Data on which calculations are based:

1. Mean forest area: 7,303 ha.
2. Mean area soya: 2,741 ha.
3. Mean area of pasture: 30,120 ha.
4. Tons of carbon per cubic meter: 0.7
5. Tons of CO2 per ton carbon: 3.67
6. Tons of carbon per Ha of forest: 200 (http://news.mongabay.com/2007/0508-amazon.html)
7. Price per head of cattle: 1.6258 R$ (000) (10% above figure in table 2)
8. Cattle per ha. 1.1 (Soares Filho, 2011)
9. Profit margin per cattle: 0.35 (From Table 2 in Somwaru, A and C. Valdes (2004)
10. Mean profit per cattle/Year: 0.6259 R$ (000). Based on above assumptions
11. SD of profit per cattle: 2.72
12. NPV of Profit from pasture: Lognormal distribution with mean 1.83 and SD 1.0
13. Price of soya: 1.4234 R$ (000) (10% above figure in table 2)
14. Yield of soya: 2.67 Ton/Ha.
http://en.mercopress.com/2012/03/20/drought-punishing-soy-crops-
and-yields-in-southern-brazil
15. Profit margin of soya 0.1 (http://www.cpac.embrapa.br/noticias/artigosmidia/publicados/359/)
16. Mean Profit from Soya/Year 0.38 R$ (000). Based on above assumptions
17. SD of profit from soya 3.0
18. NPV of profit from soya/Year Log normal distribution with mean 1.34 and SD 1.1
19. Discount rate: 10%

These are of course only just examples, and there could be many other incentives that
motivate forest clearance. Furthermore, price shocks could be much larger such that higher carbon
prices would be needed to induce the ameliorating effects that REDD may have on deforestation in
Table 4. The purpose of the calculation is to show how the result obtained here may be applied at the
municipality level to get some idea of the likely impacts of programs such as REDD. Other factors
not taken into account here are the permanence of the increase in prices or the costs of meeting the
REDD conditions. These can of course be built into the models that evaluate the impacts of such
programs.

12
7. Conclusions

This paper introduces a land-use model that measures the impact of economic and physical
variables on farmers decisions about the allocation of their land among competing uses. It is based
on a multi-output land allocation model with agricultural land as fixed and allocatable input. For the
first time, to the best of our knowledge, such a land allocation model that explicitly separates all main
competing uses for forest in Brazil soybeans, sugarcane and pasture has been estimated. By
introducing land allocation equations for sugarcane and soybean, which are not grown in many
municipalities in Brazil, we were forced to use appropriate econometric models to deal with a large
number of zeros in our dependent variables, as well as other characteristics of the data available in the
Brazilian Agricultural Census.
Our results suggest a clear substitution effect between land allocation for forest and soybeans,
and for forest and pasture. However, a limitation of our model regards the non-significant coefficients
obtained for land allocation for sugarcane; and the high variability observed for our price estimates.
A number of possibilities are foreseen to improve this land-use model in the near future and overcome
the current limitations: (i) using lagged prices for agricultural goods obtained from complementary
data sources available, which leads us to also investigate the possibility of introducing some time
dynamics to the model (a panel data analysis); (ii) introducing more relevant county-level
characteristics in the model, such as indicators of cattle intensity, historic rate of deforestation and
legal constraints (e.g. the Forest Code requirements for Legal Forest Reserve)13.
We provide two examples of the kinds of impacts we can expect from REDD programs in
private land. Additionally, simulations with our model can predict the long run land allocation among
the five uses in our model given different output prices. The coefficients estimated in our models can
also be used to estimate the shadow price of land; i.e. the price that farmers would be willing to pay
for an extra hectare of land, corresponding to their expected profit from an extra hectare of (forest)
land. These correspond to the opportunity costs of deforestation in Brazil, and can be used to derive a
REDD curve for Brazil where an important share of deforestation takes place in public land.

References
Brner, J. and Wunder, S. (2008) Paying for avoided deforestation in the Brazilian Amazon: From
cost assessment to scheme design. International Forestry Review (10)3: 496511.
Cameron, A.C. and P.K. Trivedi (2005), Microeconometrics: Methods and Applications, Cambridge
University Press, New York, NY.
Chambers, R.G. and R.E. Just (1989), Estimating multioutput technologies, American Journal of
Agricultural Economics, 71(4): 980-995.
Chomitz, K. M., P. Buys, G. De Luca, T.S. Thomas, and S. Wertz-Kanounnikoff (2007), "At
Loggerheads? Agricultural Expansion, Poverty Reduction, and Environment in the Tropical
Forests". Washington, D.C.: World Bank Research Report.
Eliasch, J. (2008) The Eliasch Review Climate Change: Financing global forests. UK Office of
Climate Change. Available online at: www.occ.gov.uk/activities/eliasch.htm
Feres, J.G. E. Reis, and J. Speranza (2010), "Climate change, land use patterns and deforestation in
Brazil", available online at

13
For example, in the Legal Amazon region the Forest Code requires that 50% of the property has to be covered
by forest.

13
 http://www.cerdi.org/uploads/sfCmsContent/html/323/Feres%20Reis.pdf, last visited
22/08/2011.
Hardie, I.W., and P.J. Parks (1997), Land use with heterogeneous land quality: An application of an
area base model, American Journal of Agricultural Economics, 79(2): 299-310.
Just, R.E., D. Zilberman, and E. Hochman (1983), Estimation of multicrop production functions,
American Journal of Agricultural Economics, 65(4): 770-780.
Lau, L.J. (1978), Applications of profit functions Production Economics: A Dual Approach to
Theory and Applications, Vol. I, eds., M. Fuss and D. McFadden, Chap. I.3. Amsterdam:
North-Holland.
Lichtenberg, E. (1989), Land quality, irrigation development, and cropping patterns in the Northern
High Plains, American Journal of Agricultural Economics, 71(1): 187-194.
Miller, D.J. and A. Plantinga (1999), Modeling land use decisions with aggregate data, American
Journal of Agricultural Economics, 81(1): 180-194.
Mckinsey & Company (2009), "Caminhos para uma Economia de Baixa Emissao de Carbono no
Brasil", available online at http://www.mckinsey.com.br/sao_paulo/carbono.pdf last visited
30/08/2011.
Moore, M.R. and A. Dinar (1995), Water and land as quantity-rationed inputs in California
agriculture: Empirical tests and water policy implications, Land Economics, 71(4): 445-461.
Moore, M.R. and D.H. Negri (1992), "A multicrop production model of irrigated agriculture, applied
to water allocation policy of the Bureau of Reclamation", Journal of Agricultural and
Resource Economics, 17(1): 29-43.
Moore, M.R., N.R. Gollehon, and M.B. Carey (1994), Production decisions in western irrigated
agriculture: The role of water price, American Journal of Agricultural Economics, 76(4):
859-874.
Nepstad, D., Soares-Filho, B. et al. (2007), "The Costs and Benefits of Reducing Carbon Emissions
from Deforestation and Forest Degradation in the Brazilian Amazon". The Woods Hole
Research Center, available online at http://www.ipam.org.br/, last visited 27/08/2011.
Olsen, N. and J. Bishop (2009), "The Financial Costs of REDD: Evidence from Brazil and Indonesia",
Gland, Switzerland: IUCN - International Union for Conservation of Nature and Natural
Resources. 64pp.
Parks, R.W. (1980), On the estimation of multinomial logit models from relative frequency data,
Journal of Econometrics, 13(3): 293-303.
Plantinga, A.J. (1996), The effect of agricultural policies on land use and environmental quality,
American Journal of Agricultural Economics, 78(4): 1082-1091.
Plantinga, A.J., T. Mauldin ,and D. Miller (1999), An econometric analysis of the costs of
sequestering carbon in forests, American Journal of Agricultural Economics, 81(4): 812-824.
Shumway, C.R. (1983), Supply, demand, and technology in a multiproduct industry: Texas field
crops, American Journal of Agricultural Economics, 65(4): 748-760.
Shumway, C.R., R.D. Pope, and E.K. Nash (1984), Allocatable fixed inputs and jointness in
agricultural production: Implications for economic modeling, American Journal of
Agricultural Economics, 66(1): 72-78.
Soares Filho, B.S. (2011) Conciliando Produo Agrcola com Conservao e Restaurao de
Florestas, available online at:
http://d3nehc6yl9qzo4.cloudfront.net/downloads/semjornalistas2011_conciliando_agricultura
_conservacao_britaldosoares.pdf
Somwaru, A and C. Valdes (2004), Brazils Beef Production and its Efficiency: a comparative study
of scale economiesavailable online at:
https://www.gtap.agecon.purdue.edu/resources/download/1860.pdf)

14
Wu, J. and Segerson, K. (1995), The impact of policies and land characteristics on potential
groundwater pollution in Wisconsin, American Journal of Agricultural Economics, 77(4):
1033-1047.

15
 BC3 WORKING PAPER SERIES
Basque Centre for Climate Change (BC3), Bilbao, Spain

The BC3 Working Paper Series is available on the internet at the following address:

http://www.bc3research.org/lits_publications.html

http://ideas.repec.org/s/bcc/wpaper.html

Recent BC3 Working Papers available:

2011-10 Aline Chiabai, Ibon Galarraga, Anil Markandya and Unai Pascual: The Equivalency
Principle for Discounting the Value of Natural Assets: An Application to an
Investment Project in the Basque Coast

2011-11 Roger Fouquet: The Demand for Environmental Quality in Driving Transitions to
Low Polluting Energy Sources

2011-12 Joseph V. Spadaro: The Uniform World Model: A Methodology for Predicting the
Health Impacts of Air Pollution

2011-13 Thomas Heyd: Climate Change, Responsibilities, and Defeatism and Complacency

2012-01 Roger Fouquet: Trends in Income and Price Elasticities of Transport Demand (1850-
2010)

2012-02 Sbastien Foudi: Exploitation of soil biota ecosystem services in agriculture: a

bioeconomic approach

2012-03 Mikel Gonzlez-Eguino, Anil Markandya and Marta Escapa: From shadow to green:
Linking environmental fiscal reform and the informal economy

2012-04 Anil Markandya, Mikel Gonzlez-Eguino and Marta Escapa: Environmental fiscal
reform and unemployment in Spain

2012-05 Dirk Rbbelke and Stefan Vgele: Effects of Carbon Dioxide Capture and Storage in
Germany on European Electricity Exchange and Welfare

2012-06 Eneko Garmendia and Gonzalo Gamboa: Weighting social preferences in

participatory multi-criteria evaluations: a case study on sustainable natural resource
management.

2012-07 Ferdinando Villa, Ken Bagstad, Gary Johnson, Brian Voigt: Towards a
Comprehensive Approach to Quantifying and Mapping Ecosystem Services

2012-08 Roger Fouquet: Economics of Energy and Climate Change: Origins, Developments
and Growth

2012-09 David R. Heres, Ramon Arigoni Ortiz and Anil Markandya: Deforestation in
Brazilian private lands: An empirical assessment of land use changes within farm